Do you trust standard errors generated by lm()?

a) Yes, always
b) Depends on the dataset and model
c) No, never
d) Not sure